fs/ceph/inode.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/slab.h>
   6 #include <linux/string.h>
   7 #include <linux/uaccess.h>
   8 #include <linux/kernel.h>
   9 #include <linux/namei.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/posix_acl.h>
  13 #include <linux/random.h>
  14
  15 #include "super.h"
  16 #include "mds_client.h"
  17 #include "cache.h"
  18 #include <linux/ceph/decode.h>
  19
  20 /*
  21  * Ceph inode operations
  22  *
  23  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  24  * setattr, etc.), xattr helpers, and helpers for assimilating
  25  * metadata returned by the MDS into our cache.
  26  *
  27  * Also define helpers for doing asynchronous writeback, invalidation,
  28  * and truncation for the benefit of those who can't afford to block
  29  * (typically because they are in the message handler path).
  30  */
  31
  32 static const struct inode_operations ceph_symlink_iops;
  33
  34 static void ceph_invalidate_work(struct work_struct *work);
  35 static void ceph_writeback_work(struct work_struct *work);
  36 static void ceph_vmtruncate_work(struct work_struct *work);
  37
  38 /*
  39  * find or create an inode, given the ceph ino number
  40  */
  41 static int ceph_set_ino_cb(struct inode *inode, void *data)
  42 {
  43         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  44         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  45         return 0;
  46 }
  47
  48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  49 {
  50         struct inode *inode;
  51         ino_t t = ceph_vino_to_ino(vino);
  52
  53         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  54         if (inode == NULL)
  55                 return ERR_PTR(-ENOMEM);
  56         if (inode->i_state & I_NEW) {
  57                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  58                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  59                 unlock_new_inode(inode);
  60         }
  61
  62         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  63              vino.snap, inode);
  64         return inode;
  65 }
  66
  67 /*
  68  * get/constuct snapdir inode for a given directory
  69  */
  70 struct inode *ceph_get_snapdir(struct inode *parent)
  71 {
  72         struct ceph_vino vino = {
  73                 .ino = ceph_ino(parent),
  74                 .snap = CEPH_SNAPDIR,
  75         };
  76         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  77         struct ceph_inode_info *ci = ceph_inode(inode);
  78
  79         BUG_ON(!S_ISDIR(parent->i_mode));
  80         if (IS_ERR(inode))
  81                 return inode;
  82         inode->i_mode = parent->i_mode;
  83         inode->i_uid = parent->i_uid;
  84         inode->i_gid = parent->i_gid;
  85         inode->i_op = &ceph_snapdir_iops;
  86         inode->i_fop = &ceph_snapdir_fops;
  87         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  88         ci->i_rbytes = 0;
  89         return inode;
  90 }
  91
  92 const struct inode_operations ceph_file_iops = {
  93         .permission = ceph_permission,
  94         .setattr = ceph_setattr,
  95         .getattr = ceph_getattr,
  96         .setxattr = ceph_setxattr,
  97         .getxattr = ceph_getxattr,
  98         .listxattr = ceph_listxattr,
  99         .removexattr = ceph_removexattr,
 100         .get_acl = ceph_get_acl,
 101         .set_acl = ceph_set_acl,
 102 };
 103
 104
 105 /*
 106  * We use a 'frag tree' to keep track of the MDS's directory fragments
 107  * for a given inode (usually there is just a single fragment).  We
 108  * need to know when a child frag is delegated to a new MDS, or when
 109  * it is flagged as replicated, so we can direct our requests
 110  * accordingly.
 111  */
 112
 113 /*
 114  * find/create a frag in the tree
 115  */
 116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 117                                                     u32 f)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct ceph_inode_frag *frag;
 122         int c;
 123
 124         p = &ci->i_fragtree.rb_node;
 125         while (*p) {
 126                 parent = *p;
 127                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 128                 c = ceph_frag_compare(f, frag->frag);
 129                 if (c < 0)
 130                         p = &(*p)->rb_left;
 131                 else if (c > 0)
 132                         p = &(*p)->rb_right;
 133                 else
 134                         return frag;
 135         }
 136
 137         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 138         if (!frag) {
 139                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 140                        "frag %x\n", &ci->vfs_inode,
 141                        ceph_vinop(&ci->vfs_inode), f);
 142                 return ERR_PTR(-ENOMEM);
 143         }
 144         frag->frag = f;
 145         frag->split_by = 0;
 146         frag->mds = -1;
 147         frag->ndist = 0;
 148
 149         rb_link_node(&frag->node, parent, p);
 150         rb_insert_color(&frag->node, &ci->i_fragtree);
 151
 152         dout("get_or_create_frag added %llx.%llx frag %x\n",
 153              ceph_vinop(&ci->vfs_inode), f);
 154         return frag;
 155 }
 156
 157 /*
 158  * find a specific frag @f
 159  */
 160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 161 {
 162         struct rb_node *n = ci->i_fragtree.rb_node;
 163
 164         while (n) {
 165                 struct ceph_inode_frag *frag =
 166                         rb_entry(n, struct ceph_inode_frag, node);
 167                 int c = ceph_frag_compare(f, frag->frag);
 168                 if (c < 0)
 169                         n = n->rb_left;
 170                 else if (c > 0)
 171                         n = n->rb_right;
 172                 else
 173                         return frag;
 174         }
 175         return NULL;
 176 }
 177
 178 /*
 179  * Choose frag containing the given value @v.  If @pfrag is
 180  * specified, copy the frag delegation info to the caller if
 181  * it is present.
 182  */
 183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 184                               struct ceph_inode_frag *pfrag, int *found)
 185 {
 186         u32 t = ceph_frag_make(0, 0);
 187         struct ceph_inode_frag *frag;
 188         unsigned nway, i;
 189         u32 n;
 190
 191         if (found)
 192                 *found = 0;
 193
 194         while (1) {
 195                 WARN_ON(!ceph_frag_contains_value(t, v));
 196                 frag = __ceph_find_frag(ci, t);
 197                 if (!frag)
 198                         break; /* t is a leaf */
 199                 if (frag->split_by == 0) {
 200                         if (pfrag)
 201                                 memcpy(pfrag, frag, sizeof(*pfrag));
 202                         if (found)
 203                                 *found = 1;
 204                         break;
 205                 }
 206
 207                 /* choose child */
 208                 nway = 1 << frag->split_by;
 209                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 210                      frag->split_by, nway);
 211                 for (i = 0; i < nway; i++) {
 212                         n = ceph_frag_make_child(t, frag->split_by, i);
 213                         if (ceph_frag_contains_value(n, v)) {
 214                                 t = n;
 215                                 break;
 216                         }
 217                 }
 218                 BUG_ON(i == nway);
 219         }
 220         dout("choose_frag(%x) = %x\n", v, t);
 221
 222         return t;
 223 }
 224
 225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 226                      struct ceph_inode_frag *pfrag, int *found)
 227 {
 228         u32 ret;
 229         mutex_lock(&ci->i_fragtree_mutex);
 230         ret = __ceph_choose_frag(ci, v, pfrag, found);
 231         mutex_unlock(&ci->i_fragtree_mutex);
 232         return ret;
 233 }
 234
 235 /*
 236  * Process dirfrag (delegation) info from the mds.  Include leaf
 237  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 238  * branches/splits are included in i_fragtree)
 239  */
 240 static int ceph_fill_dirfrag(struct inode *inode,
 241                              struct ceph_mds_reply_dirfrag *dirinfo)
 242 {
 243         struct ceph_inode_info *ci = ceph_inode(inode);
 244         struct ceph_inode_frag *frag;
 245         u32 id = le32_to_cpu(dirinfo->frag);
 246         int mds = le32_to_cpu(dirinfo->auth);
 247         int ndist = le32_to_cpu(dirinfo->ndist);
 248         int diri_auth = -1;
 249         int i;
 250         int err = 0;
 251
 252         spin_lock(&ci->i_ceph_lock);
 253         if (ci->i_auth_cap)
 254                 diri_auth = ci->i_auth_cap->mds;
 255         spin_unlock(&ci->i_ceph_lock);
 256
 257         mutex_lock(&ci->i_fragtree_mutex);
 258         if (ndist == 0 && mds == diri_auth) {
 259                 /* no delegation info needed. */
 260                 frag = __ceph_find_frag(ci, id);
 261                 if (!frag)
 262                         goto out;
 263                 if (frag->split_by == 0) {
 264                         /* tree leaf, remove */
 265                         dout("fill_dirfrag removed %llx.%llx frag %x"
 266                              " (no ref)\n", ceph_vinop(inode), id);
 267                         rb_erase(&frag->node, &ci->i_fragtree);
 268                         kfree(frag);
 269                 } else {
 270                         /* tree branch, keep and clear */
 271                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 272                              " referral\n", ceph_vinop(inode), id);
 273                         frag->mds = -1;
 274                         frag->ndist = 0;
 275                 }
 276                 goto out;
 277         }
 278
 279
 280         /* find/add this frag to store mds delegation info */
 281         frag = __get_or_create_frag(ci, id);
 282         if (IS_ERR(frag)) {
 283                 /* this is not the end of the world; we can continue
 284                    with bad/inaccurate delegation info */
 285                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 286                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 287                 err = -ENOMEM;
 288                 goto out;
 289         }
 290
 291         frag->mds = mds;
 292         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 293         for (i = 0; i < frag->ndist; i++)
 294                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 295         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 296              ceph_vinop(inode), frag->frag, frag->ndist);
 297
 298 out:
 299         mutex_unlock(&ci->i_fragtree_mutex);
 300         return err;
 301 }
 302
 303 static int ceph_fill_fragtree(struct inode *inode,
 304                               struct ceph_frag_tree_head *fragtree,
 305                               struct ceph_mds_reply_dirfrag *dirinfo)
 306 {
 307         struct ceph_inode_info *ci = ceph_inode(inode);
 308         struct ceph_inode_frag *frag;
 309         struct rb_node *rb_node;
 310         int i;
 311         u32 id, nsplits;
 312         bool update = false;
 313
 314         mutex_lock(&ci->i_fragtree_mutex);
 315         nsplits = le32_to_cpu(fragtree->nsplits);
 316         if (nsplits) {
 317                 i = prandom_u32() % nsplits;
 318                 id = le32_to_cpu(fragtree->splits[i].frag);
 319                 if (!__ceph_find_frag(ci, id))
 320                         update = true;
 321         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 322                 rb_node = rb_first(&ci->i_fragtree);
 323                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 324                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 325                         update = true;
 326         }
 327         if (!update && dirinfo) {
 328                 id = le32_to_cpu(dirinfo->frag);
 329                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 330                         update = true;
 331         }
 332         if (!update)
 333                 goto out_unlock;
 334
 335         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 336         rb_node = rb_first(&ci->i_fragtree);
 337         for (i = 0; i < nsplits; i++) {
 338                 id = le32_to_cpu(fragtree->splits[i].frag);
 339                 frag = NULL;
 340                 while (rb_node) {
 341                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 343                                 if (frag->frag != id)
 344                                         frag = NULL;
 345                                 else
 346                                         rb_node = rb_next(rb_node);
 347                                 break;
 348                         }
 349                         rb_node = rb_next(rb_node);
 350                         rb_erase(&frag->node, &ci->i_fragtree);
 351                         kfree(frag);
 352                         frag = NULL;
 353                 }
 354                 if (!frag) {
 355                         frag = __get_or_create_frag(ci, id);
 356                         if (IS_ERR(frag))
 357                                 continue;
 358                 }
 359                 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
 360                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 361         }
 362         while (rb_node) {
 363                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 364                 rb_node = rb_next(rb_node);
 365                 rb_erase(&frag->node, &ci->i_fragtree);
 366                 kfree(frag);
 367         }
 368 out_unlock:
 369         mutex_unlock(&ci->i_fragtree_mutex);
 370         return 0;
 371 }
 372
 373 /*
 374  * initialize a newly allocated inode.
 375  */
 376 struct inode *ceph_alloc_inode(struct super_block *sb)
 377 {
 378         struct ceph_inode_info *ci;
 379         int i;
 380
 381         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 382         if (!ci)
 383                 return NULL;
 384
 385         dout("alloc_inode %p\n", &ci->vfs_inode);
 386
 387         spin_lock_init(&ci->i_ceph_lock);
 388
 389         ci->i_version = 0;
 390         ci->i_inline_version = 0;
 391         ci->i_time_warp_seq = 0;
 392         ci->i_ceph_flags = 0;
 393         ci->i_ordered_count = 0;
 394         atomic_set(&ci->i_release_count, 1);
 395         atomic_set(&ci->i_complete_count, 0);
 396         ci->i_symlink = NULL;
 397
 398         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 399
 400         ci->i_fragtree = RB_ROOT;
 401         mutex_init(&ci->i_fragtree_mutex);
 402
 403         ci->i_xattrs.blob = NULL;
 404         ci->i_xattrs.prealloc_blob = NULL;
 405         ci->i_xattrs.dirty = false;
 406         ci->i_xattrs.index = RB_ROOT;
 407         ci->i_xattrs.count = 0;
 408         ci->i_xattrs.names_size = 0;
 409         ci->i_xattrs.vals_size = 0;
 410         ci->i_xattrs.version = 0;
 411         ci->i_xattrs.index_version = 0;
 412
 413         ci->i_caps = RB_ROOT;
 414         ci->i_auth_cap = NULL;
 415         ci->i_dirty_caps = 0;
 416         ci->i_flushing_caps = 0;
 417         INIT_LIST_HEAD(&ci->i_dirty_item);
 418         INIT_LIST_HEAD(&ci->i_flushing_item);
 419         ci->i_cap_flush_seq = 0;
 420         ci->i_cap_flush_last_tid = 0;
 421         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 422         init_waitqueue_head(&ci->i_cap_wq);
 423         ci->i_hold_caps_min = 0;
 424         ci->i_hold_caps_max = 0;
 425         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 426         INIT_LIST_HEAD(&ci->i_cap_snaps);
 427         ci->i_head_snapc = NULL;
 428         ci->i_snap_caps = 0;
 429
 430         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 431                 ci->i_nr_by_mode[i] = 0;
 432
 433         mutex_init(&ci->i_truncate_mutex);
 434         ci->i_truncate_seq = 0;
 435         ci->i_truncate_size = 0;
 436         ci->i_truncate_pending = 0;
 437
 438         ci->i_max_size = 0;
 439         ci->i_reported_size = 0;
 440         ci->i_wanted_max_size = 0;
 441         ci->i_requested_max_size = 0;
 442
 443         ci->i_pin_ref = 0;
 444         ci->i_rd_ref = 0;
 445         ci->i_rdcache_ref = 0;
 446         ci->i_wr_ref = 0;
 447         ci->i_wb_ref = 0;
 448         ci->i_wrbuffer_ref = 0;
 449         ci->i_wrbuffer_ref_head = 0;
 450         ci->i_shared_gen = 0;
 451         ci->i_rdcache_gen = 0;
 452         ci->i_rdcache_revoking = 0;
 453
 454         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 455         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 456         spin_lock_init(&ci->i_unsafe_lock);
 457
 458         ci->i_snap_realm = NULL;
 459         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 460         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 461
 462         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 463         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 464
 465         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 466
 467         ceph_fscache_inode_init(ci);
 468
 469         return &ci->vfs_inode;
 470 }
 471
 472 static void ceph_i_callback(struct rcu_head *head)
 473 {
 474         struct inode *inode = container_of(head, struct inode, i_rcu);
 475         struct ceph_inode_info *ci = ceph_inode(inode);
 476
 477         kmem_cache_free(ceph_inode_cachep, ci);
 478 }
 479
 480 void ceph_destroy_inode(struct inode *inode)
 481 {
 482         struct ceph_inode_info *ci = ceph_inode(inode);
 483         struct ceph_inode_frag *frag;
 484         struct rb_node *n;
 485
 486         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 487
 488         ceph_fscache_unregister_inode_cookie(ci);
 489
 490         ceph_queue_caps_release(inode);
 491
 492         /*
 493          * we may still have a snap_realm reference if there are stray
 494          * caps in i_snap_caps.
 495          */
 496         if (ci->i_snap_realm) {
 497                 struct ceph_mds_client *mdsc =
 498                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 499                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 500
 501                 dout(" dropping residual ref to snap realm %p\n", realm);
 502                 spin_lock(&realm->inodes_with_caps_lock);
 503                 list_del_init(&ci->i_snap_realm_item);
 504                 spin_unlock(&realm->inodes_with_caps_lock);
 505                 ceph_put_snap_realm(mdsc, realm);
 506         }
 507
 508         kfree(ci->i_symlink);
 509         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 510                 frag = rb_entry(n, struct ceph_inode_frag, node);
 511                 rb_erase(n, &ci->i_fragtree);
 512                 kfree(frag);
 513         }
 514
 515         __ceph_destroy_xattrs(ci);
 516         if (ci->i_xattrs.blob)
 517                 ceph_buffer_put(ci->i_xattrs.blob);
 518         if (ci->i_xattrs.prealloc_blob)
 519                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 520
 521         call_rcu(&inode->i_rcu, ceph_i_callback);
 522 }
 523
 524 int ceph_drop_inode(struct inode *inode)
 525 {
 526         /*
 527          * Positve dentry and corresponding inode are always accompanied
 528          * in MDS reply. So no need to keep inode in the cache after
 529          * dropping all its aliases.
 530          */
 531         return 1;
 532 }
 533
 534 /*
 535  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 536  * careful because either the client or MDS may have more up to date
 537  * info, depending on which capabilities are held, and whether
 538  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 539  * and size are monotonically increasing, except when utimes() or
 540  * truncate() increments the corresponding _seq values.)
 541  */
 542 int ceph_fill_file_size(struct inode *inode, int issued,
 543                         u32 truncate_seq, u64 truncate_size, u64 size)
 544 {
 545         struct ceph_inode_info *ci = ceph_inode(inode);
 546         int queue_trunc = 0;
 547
 548         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 549             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 550                 dout("size %lld -> %llu\n", inode->i_size, size);
 551                 inode->i_size = size;
 552                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 553                 ci->i_reported_size = size;
 554                 if (truncate_seq != ci->i_truncate_seq) {
 555                         dout("truncate_seq %u -> %u\n",
 556                              ci->i_truncate_seq, truncate_seq);
 557                         ci->i_truncate_seq = truncate_seq;
 558
 559                         /* the MDS should have revoked these caps */
 560                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 561                                                CEPH_CAP_FILE_RD |
 562                                                CEPH_CAP_FILE_WR |
 563                                                CEPH_CAP_FILE_LAZYIO));
 564                         /*
 565                          * If we hold relevant caps, or in the case where we're
 566                          * not the only client referencing this file and we
 567                          * don't hold those caps, then we need to check whether
 568                          * the file is either opened or mmaped
 569                          */
 570                         if ((issued & (CEPH_CAP_FILE_CACHE|
 571                                        CEPH_CAP_FILE_BUFFER)) ||
 572                             mapping_mapped(inode->i_mapping) ||
 573                             __ceph_caps_file_wanted(ci)) {
 574                                 ci->i_truncate_pending++;
 575                                 queue_trunc = 1;
 576                         }
 577                 }
 578         }
 579         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 580             ci->i_truncate_size != truncate_size) {
 581                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 582                      truncate_size);
 583                 ci->i_truncate_size = truncate_size;
 584         }
 585
 586         if (queue_trunc)
 587                 ceph_fscache_invalidate(inode);
 588
 589         return queue_trunc;
 590 }
 591
 592 void ceph_fill_file_time(struct inode *inode, int issued,
 593                          u64 time_warp_seq, struct timespec *ctime,
 594                          struct timespec *mtime, struct timespec *atime)
 595 {
 596         struct ceph_inode_info *ci = ceph_inode(inode);
 597         int warn = 0;
 598
 599         if (issued & (CEPH_CAP_FILE_EXCL|
 600                       CEPH_CAP_FILE_WR|
 601                       CEPH_CAP_FILE_BUFFER|
 602                       CEPH_CAP_AUTH_EXCL|
 603                       CEPH_CAP_XATTR_EXCL)) {
 604                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 605                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 606                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 607                              ctime->tv_sec, ctime->tv_nsec);
 608                         inode->i_ctime = *ctime;
 609                 }
 610                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 611                         /* the MDS did a utimes() */
 612                         dout("mtime %ld.%09ld -> %ld.%09ld "
 613                              "tw %d -> %d\n",
 614                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 615                              mtime->tv_sec, mtime->tv_nsec,
 616                              ci->i_time_warp_seq, (int)time_warp_seq);
 617
 618                         inode->i_mtime = *mtime;
 619                         inode->i_atime = *atime;
 620                         ci->i_time_warp_seq = time_warp_seq;
 621                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 622                         /* nobody did utimes(); take the max */
 623                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 624                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 625                                      inode->i_mtime.tv_sec,
 626                                      inode->i_mtime.tv_nsec,
 627                                      mtime->tv_sec, mtime->tv_nsec);
 628                                 inode->i_mtime = *mtime;
 629                         }
 630                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 631                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 632                                      inode->i_atime.tv_sec,
 633                                      inode->i_atime.tv_nsec,
 634                                      atime->tv_sec, atime->tv_nsec);
 635                                 inode->i_atime = *atime;
 636                         }
 637                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 638                         /* we did a utimes(); ignore mds values */
 639                 } else {
 640                         warn = 1;
 641                 }
 642         } else {
 643                 /* we have no write|excl caps; whatever the MDS says is true */
 644                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 645                         inode->i_ctime = *ctime;
 646                         inode->i_mtime = *mtime;
 647                         inode->i_atime = *atime;
 648                         ci->i_time_warp_seq = time_warp_seq;
 649                 } else {
 650                         warn = 1;
 651                 }
 652         }
 653         if (warn) /* time_warp_seq shouldn't go backwards */
 654                 dout("%p mds time_warp_seq %llu < %u\n",
 655                      inode, time_warp_seq, ci->i_time_warp_seq);
 656 }
 657
 658 /*
 659  * Populate an inode based on info from mds.  May be called on new or
 660  * existing inodes.
 661  */
 662 static int fill_inode(struct inode *inode, struct page *locked_page,
 663                       struct ceph_mds_reply_info_in *iinfo,
 664                       struct ceph_mds_reply_dirfrag *dirinfo,
 665                       struct ceph_mds_session *session,
 666                       unsigned long ttl_from, int cap_fmode,
 667                       struct ceph_cap_reservation *caps_reservation)
 668 {
 669         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 670         struct ceph_mds_reply_inode *info = iinfo->in;
 671         struct ceph_inode_info *ci = ceph_inode(inode);
 672         int issued = 0, implemented, new_issued;
 673         struct timespec mtime, atime, ctime;
 674         struct ceph_buffer *xattr_blob = NULL;
 675         struct ceph_cap *new_cap = NULL;
 676         int err = 0;
 677         bool wake = false;
 678         bool queue_trunc = false;
 679         bool new_version = false;
 680         bool fill_inline = false;
 681
 682         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 683              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 684              ci->i_version);
 685
 686         /* prealloc new cap struct */
 687         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 688                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 689
 690         /*
 691          * prealloc xattr data, if it looks like we'll need it.  only
 692          * if len > 4 (meaning there are actually xattrs; the first 4
 693          * bytes are the xattr count).
 694          */
 695         if (iinfo->xattr_len > 4) {
 696                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 697                 if (!xattr_blob)
 698                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 699                                iinfo->xattr_len);
 700         }
 701
 702         spin_lock(&ci->i_ceph_lock);
 703
 704         /*
 705          * provided version will be odd if inode value is projected,
 706          * even if stable.  skip the update if we have newer stable
 707          * info (ours>=theirs, e.g. due to racing mds replies), unless
 708          * we are getting projected (unstable) info (in which case the
 709          * version is odd, and we want ours>theirs).
 710          *   us   them
 711          *   2    2     skip
 712          *   3    2     skip
 713          *   3    3     update
 714          */
 715         if (ci->i_version == 0 ||
 716             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 717              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 718                 new_version = true;
 719
 720         issued = __ceph_caps_issued(ci, &implemented);
 721         issued |= implemented | __ceph_caps_dirty(ci);
 722         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 723
 724         /* update inode */
 725         ci->i_version = le64_to_cpu(info->version);
 726         inode->i_version++;
 727         inode->i_rdev = le32_to_cpu(info->rdev);
 728         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 729
 730         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 731             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 732                 inode->i_mode = le32_to_cpu(info->mode);
 733                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 734                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 735                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 736                      from_kuid(&init_user_ns, inode->i_uid),
 737                      from_kgid(&init_user_ns, inode->i_gid));
 738         }
 739
 740         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 741             (issued & CEPH_CAP_LINK_EXCL) == 0)
 742                 set_nlink(inode, le32_to_cpu(info->nlink));
 743
 744         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 745                 /* be careful with mtime, atime, size */
 746                 ceph_decode_timespec(&atime, &info->atime);
 747                 ceph_decode_timespec(&mtime, &info->mtime);
 748                 ceph_decode_timespec(&ctime, &info->ctime);
 749                 ceph_fill_file_time(inode, issued,
 750                                 le32_to_cpu(info->time_warp_seq),
 751                                 &ctime, &mtime, &atime);
 752         }
 753
 754         if (new_version ||
 755             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 756                 ci->i_layout = info->layout;
 757                 queue_trunc = ceph_fill_file_size(inode, issued,
 758                                         le32_to_cpu(info->truncate_seq),
 759                                         le64_to_cpu(info->truncate_size),
 760                                         le64_to_cpu(info->size));
 761                 /* only update max_size on auth cap */
 762                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 763                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 764                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 765                                         le64_to_cpu(info->max_size));
 766                         ci->i_max_size = le64_to_cpu(info->max_size);
 767                 }
 768         }
 769
 770         /* xattrs */
 771         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 772         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 773             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 774                 if (ci->i_xattrs.blob)
 775                         ceph_buffer_put(ci->i_xattrs.blob);
 776                 ci->i_xattrs.blob = xattr_blob;
 777                 if (xattr_blob)
 778                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 779                                iinfo->xattr_data, iinfo->xattr_len);
 780                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 781                 ceph_forget_all_cached_acls(inode);
 782                 xattr_blob = NULL;
 783         }
 784
 785         inode->i_mapping->a_ops = &ceph_aops;
 786         inode->i_mapping->backing_dev_info =
 787                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
 788
 789         switch (inode->i_mode & S_IFMT) {
 790         case S_IFIFO:
 791         case S_IFBLK:
 792         case S_IFCHR:
 793         case S_IFSOCK:
 794                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 795                 inode->i_op = &ceph_file_iops;
 796                 break;
 797         case S_IFREG:
 798                 inode->i_op = &ceph_file_iops;
 799                 inode->i_fop = &ceph_file_fops;
 800                 break;
 801         case S_IFLNK:
 802                 inode->i_op = &ceph_symlink_iops;
 803                 if (!ci->i_symlink) {
 804                         u32 symlen = iinfo->symlink_len;
 805                         char *sym;
 806
 807                         spin_unlock(&ci->i_ceph_lock);
 808
 809                         err = -EINVAL;
 810                         if (WARN_ON(symlen != inode->i_size))
 811                                 goto out;
 812
 813                         err = -ENOMEM;
 814                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 815                         if (!sym)
 816                                 goto out;
 817
 818                         spin_lock(&ci->i_ceph_lock);
 819                         if (!ci->i_symlink)
 820                                 ci->i_symlink = sym;
 821                         else
 822                                 kfree(sym); /* lost a race */
 823                 }
 824                 break;
 825         case S_IFDIR:
 826                 inode->i_op = &ceph_dir_iops;
 827                 inode->i_fop = &ceph_dir_fops;
 828
 829                 ci->i_dir_layout = iinfo->dir_layout;
 830
 831                 ci->i_files = le64_to_cpu(info->files);
 832                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 833                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 834                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 835                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 836                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 837                 break;
 838         default:
 839                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 840                        ceph_vinop(inode), inode->i_mode);
 841         }
 842
 843         /* were we issued a capability? */
 844         if (info->cap.caps) {
 845                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 846                         unsigned caps = le32_to_cpu(info->cap.caps);
 847                         ceph_add_cap(inode, session,
 848                                      le64_to_cpu(info->cap.cap_id),
 849                                      cap_fmode, caps,
 850                                      le32_to_cpu(info->cap.wanted),
 851                                      le32_to_cpu(info->cap.seq),
 852                                      le32_to_cpu(info->cap.mseq),
 853                                      le64_to_cpu(info->cap.realm),
 854                                      info->cap.flags, &new_cap);
 855
 856                         /* set dir completion flag? */
 857                         if (S_ISDIR(inode->i_mode) &&
 858                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 859                             (caps & CEPH_CAP_FILE_SHARED) &&
 860                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 861                             !__ceph_dir_is_complete(ci)) {
 862                                 dout(" marking %p complete (empty)\n", inode);
 863                                 __ceph_dir_set_complete(ci,
 864                                         atomic_read(&ci->i_release_count),
 865                                         ci->i_ordered_count);
 866                         }
 867
 868                         wake = true;
 869                 } else {
 870                         dout(" %p got snap_caps %s\n", inode,
 871                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 872                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 873                         if (cap_fmode >= 0)
 874                                 __ceph_get_fmode(ci, cap_fmode);
 875                 }
 876         } else if (cap_fmode >= 0) {
 877                 pr_warn("mds issued no caps on %llx.%llx\n",
 878                            ceph_vinop(inode));
 879                 __ceph_get_fmode(ci, cap_fmode);
 880         }
 881
 882         if (iinfo->inline_version > 0 &&
 883             iinfo->inline_version >= ci->i_inline_version) {
 884                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 885                 ci->i_inline_version = iinfo->inline_version;
 886                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 887                     (locked_page ||
 888                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 889                         fill_inline = true;
 890         }
 891
 892         spin_unlock(&ci->i_ceph_lock);
 893
 894         if (fill_inline)
 895                 ceph_fill_inline_data(inode, locked_page,
 896                                       iinfo->inline_data, iinfo->inline_len);
 897
 898         if (wake)
 899                 wake_up_all(&ci->i_cap_wq);
 900
 901         /* queue truncate if we saw i_size decrease */
 902         if (queue_trunc)
 903                 ceph_queue_vmtruncate(inode);
 904
 905         /* populate frag tree */
 906         if (S_ISDIR(inode->i_mode))
 907                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 908
 909         /* update delegation info? */
 910         if (dirinfo)
 911                 ceph_fill_dirfrag(inode, dirinfo);
 912
 913         err = 0;
 914 out:
 915         if (new_cap)
 916                 ceph_put_cap(mdsc, new_cap);
 917         if (xattr_blob)
 918                 ceph_buffer_put(xattr_blob);
 919         return err;
 920 }
 921
 922 /*
 923  * caller should hold session s_mutex.
 924  */
 925 static void update_dentry_lease(struct dentry *dentry,
 926                                 struct ceph_mds_reply_lease *lease,
 927                                 struct ceph_mds_session *session,
 928                                 unsigned long from_time)
 929 {
 930         struct ceph_dentry_info *di = ceph_dentry(dentry);
 931         long unsigned duration = le32_to_cpu(lease->duration_ms);
 932         long unsigned ttl = from_time + (duration * HZ) / 1000;
 933         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 934         struct inode *dir;
 935
 936         /* only track leases on regular dentries */
 937         if (dentry->d_op != &ceph_dentry_ops)
 938                 return;
 939
 940         spin_lock(&dentry->d_lock);
 941         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 942              dentry, duration, ttl);
 943
 944         /* make lease_rdcache_gen match directory */
 945         dir = dentry->d_parent->d_inode;
 946         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 947
 948         if (duration == 0)
 949                 goto out_unlock;
 950
 951         if (di->lease_gen == session->s_cap_gen &&
 952             time_before(ttl, dentry->d_time))
 953                 goto out_unlock;  /* we already have a newer lease. */
 954
 955         if (di->lease_session && di->lease_session != session)
 956                 goto out_unlock;
 957
 958         ceph_dentry_lru_touch(dentry);
 959
 960         if (!di->lease_session)
 961                 di->lease_session = ceph_get_mds_session(session);
 962         di->lease_gen = session->s_cap_gen;
 963         di->lease_seq = le32_to_cpu(lease->seq);
 964         di->lease_renew_after = half_ttl;
 965         di->lease_renew_from = 0;
 966         dentry->d_time = ttl;
 967 out_unlock:
 968         spin_unlock(&dentry->d_lock);
 969         return;
 970 }
 971
 972 /*
 973  * splice a dentry to an inode.
 974  * caller must hold directory i_mutex for this to be safe.
 975  *
 976  * we will only rehash the resulting dentry if @prehash is
 977  * true; @prehash will be set to false (for the benefit of
 978  * the caller) if we fail.
 979  */
 980 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 981                                     bool *prehash)
 982 {
 983         struct dentry *realdn;
 984
 985         BUG_ON(dn->d_inode);
 986
 987         /* dn must be unhashed */
 988         if (!d_unhashed(dn))
 989                 d_drop(dn);
 990         realdn = d_splice_alias(in, dn);
 991         if (IS_ERR(realdn)) {
 992                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 993                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 994                 if (prehash)
 995                         *prehash = false; /* don't rehash on error */
 996                 dn = realdn; /* note realdn contains the error */
 997                 goto out;
 998         } else if (realdn) {
 999                 dout("dn %p (%d) spliced with %p (%d) "
1000                      "inode %p ino %llx.%llx\n",
1001                      dn, d_count(dn),
1002                      realdn, d_count(realdn),
1003                      realdn->d_inode, ceph_vinop(realdn->d_inode));
1004                 dput(dn);
1005                 dn = realdn;
1006         } else {
1007                 BUG_ON(!ceph_dentry(dn));
1008                 dout("dn %p attached to %p ino %llx.%llx\n",
1009                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
1010         }
1011         if ((!prehash || *prehash) && d_unhashed(dn))
1012                 d_rehash(dn);
1013 out:
1014         return dn;
1015 }
1016
1017 /*
1018  * Incorporate results into the local cache.  This is either just
1019  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1020  * after a lookup).
1021  *
1022  * A reply may contain
1023  *         a directory inode along with a dentry.
1024  *  and/or a target inode
1025  *
1026  * Called with snap_rwsem (read).
1027  */
1028 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1029                     struct ceph_mds_session *session)
1030 {
1031         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1032         struct inode *in = NULL;
1033         struct ceph_vino vino;
1034         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1035         int err = 0;
1036
1037         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1038              rinfo->head->is_dentry, rinfo->head->is_target);
1039
1040 #if 0
1041         /*
1042          * Debugging hook:
1043          *
1044          * If we resend completed ops to a recovering mds, we get no
1045          * trace.  Since that is very rare, pretend this is the case
1046          * to ensure the 'no trace' handlers in the callers behave.
1047          *
1048          * Fill in inodes unconditionally to avoid breaking cap
1049          * invariants.
1050          */
1051         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1052                 pr_info("fill_trace faking empty trace on %lld %s\n",
1053                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1054                 if (rinfo->head->is_dentry) {
1055                         rinfo->head->is_dentry = 0;
1056                         err = fill_inode(req->r_locked_dir,
1057                                          &rinfo->diri, rinfo->dirfrag,
1058                                          session, req->r_request_started, -1);
1059                 }
1060                 if (rinfo->head->is_target) {
1061                         rinfo->head->is_target = 0;
1062                         ininfo = rinfo->targeti.in;
1063                         vino.ino = le64_to_cpu(ininfo->ino);
1064                         vino.snap = le64_to_cpu(ininfo->snapid);
1065                         in = ceph_get_inode(sb, vino);
1066                         err = fill_inode(in, &rinfo->targeti, NULL,
1067                                          session, req->r_request_started,
1068                                          req->r_fmode);
1069                         iput(in);
1070                 }
1071         }
1072 #endif
1073
1074         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1075                 dout("fill_trace reply is empty!\n");
1076                 if (rinfo->head->result == 0 && req->r_locked_dir)
1077                         ceph_invalidate_dir_request(req);
1078                 return 0;
1079         }
1080
1081         if (rinfo->head->is_dentry) {
1082                 struct inode *dir = req->r_locked_dir;
1083
1084                 if (dir) {
1085                         err = fill_inode(dir, NULL,
1086                                          &rinfo->diri, rinfo->dirfrag,
1087                                          session, req->r_request_started, -1,
1088                                          &req->r_caps_reservation);
1089                         if (err < 0)
1090                                 goto done;
1091                 } else {
1092                         WARN_ON_ONCE(1);
1093                 }
1094
1095                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1096                         struct qstr dname;
1097                         struct dentry *dn, *parent;
1098
1099                         BUG_ON(!rinfo->head->is_target);
1100                         BUG_ON(req->r_dentry);
1101
1102                         parent = d_find_any_alias(dir);
1103                         BUG_ON(!parent);
1104
1105                         dname.name = rinfo->dname;
1106                         dname.len = rinfo->dname_len;
1107                         dname.hash = full_name_hash(dname.name, dname.len);
1108                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1109                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1110 retry_lookup:
1111                         dn = d_lookup(parent, &dname);
1112                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1113                              parent, dname.len, dname.name, dn);
1114
1115                         if (!dn) {
1116                                 dn = d_alloc(parent, &dname);
1117                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1118                                      dname.len, dname.name, dn);
1119                                 if (dn == NULL) {
1120                                         dput(parent);
1121                                         err = -ENOMEM;
1122                                         goto done;
1123                                 }
1124                                 err = ceph_init_dentry(dn);
1125                                 if (err < 0) {
1126                                         dput(dn);
1127                                         dput(parent);
1128                                         goto done;
1129                                 }
1130                         } else if (dn->d_inode &&
1131                                    (ceph_ino(dn->d_inode) != vino.ino ||
1132                                     ceph_snap(dn->d_inode) != vino.snap)) {
1133                                 dout(" dn %p points to wrong inode %p\n",
1134                                      dn, dn->d_inode);
1135                                 d_delete(dn);
1136                                 dput(dn);
1137                                 goto retry_lookup;
1138                         }
1139
1140                         req->r_dentry = dn;
1141                         dput(parent);
1142                 }
1143         }
1144
1145         if (rinfo->head->is_target) {
1146                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1147                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1148
1149                 in = ceph_get_inode(sb, vino);
1150                 if (IS_ERR(in)) {
1151                         err = PTR_ERR(in);
1152                         goto done;
1153                 }
1154                 req->r_target_inode = in;
1155
1156                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1157                                 session, req->r_request_started,
1158                                 (!req->r_aborted && rinfo->head->result == 0) ?
1159                                 req->r_fmode : -1,
1160                                 &req->r_caps_reservation);
1161                 if (err < 0) {
1162                         pr_err("fill_inode badness %p %llx.%llx\n",
1163                                 in, ceph_vinop(in));
1164                         goto done;
1165                 }
1166         }
1167
1168         /*
1169          * ignore null lease/binding on snapdir ENOENT, or else we
1170          * will have trouble splicing in the virtual snapdir later
1171          */
1172         if (rinfo->head->is_dentry && !req->r_aborted &&
1173             req->r_locked_dir &&
1174             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1175                                                fsc->mount_options->snapdir_name,
1176                                                req->r_dentry->d_name.len))) {
1177                 /*
1178                  * lookup link rename   : null -> possibly existing inode
1179                  * mknod symlink mkdir  : null -> new inode
1180                  * unlink               : linked -> null
1181                  */
1182                 struct inode *dir = req->r_locked_dir;
1183                 struct dentry *dn = req->r_dentry;
1184                 bool have_dir_cap, have_lease;
1185
1186                 BUG_ON(!dn);
1187                 BUG_ON(!dir);
1188                 BUG_ON(dn->d_parent->d_inode != dir);
1189                 BUG_ON(ceph_ino(dir) !=
1190                        le64_to_cpu(rinfo->diri.in->ino));
1191                 BUG_ON(ceph_snap(dir) !=
1192                        le64_to_cpu(rinfo->diri.in->snapid));
1193
1194                 /* do we have a lease on the whole dir? */
1195                 have_dir_cap =
1196                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1197                          CEPH_CAP_FILE_SHARED);
1198
1199                 /* do we have a dn lease? */
1200                 have_lease = have_dir_cap ||
1201                         le32_to_cpu(rinfo->dlease->duration_ms);
1202                 if (!have_lease)
1203                         dout("fill_trace  no dentry lease or dir cap\n");
1204
1205                 /* rename? */
1206                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1207                         struct inode *olddir = req->r_old_dentry_dir;
1208                         BUG_ON(!olddir);
1209
1210                         dout(" src %p '%pd' dst %p '%pd'\n",
1211                              req->r_old_dentry,
1212                              req->r_old_dentry,
1213                              dn, dn);
1214                         dout("fill_trace doing d_move %p -> %p\n",
1215                              req->r_old_dentry, dn);
1216
1217                         d_move(req->r_old_dentry, dn);
1218                         dout(" src %p '%pd' dst %p '%pd'\n",
1219                              req->r_old_dentry,
1220                              req->r_old_dentry,
1221                              dn, dn);
1222
1223                         /* ensure target dentry is invalidated, despite
1224                            rehashing bug in vfs_rename_dir */
1225                         ceph_invalidate_dentry_lease(dn);
1226
1227                         /* d_move screws up sibling dentries' offsets */
1228                         ceph_dir_clear_ordered(dir);
1229                         ceph_dir_clear_ordered(olddir);
1230
1231                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1232                              ceph_dentry(req->r_old_dentry)->offset);
1233
1234                         dn = req->r_old_dentry;  /* use old_dentry */
1235                 }
1236
1237                 /* null dentry? */
1238                 if (!rinfo->head->is_target) {
1239                         dout("fill_trace null dentry\n");
1240                         if (dn->d_inode) {
1241                                 ceph_dir_clear_ordered(dir);
1242                                 dout("d_delete %p\n", dn);
1243                                 d_delete(dn);
1244                         } else {
1245                                 dout("d_instantiate %p NULL\n", dn);
1246                                 d_instantiate(dn, NULL);
1247                                 if (have_lease && d_unhashed(dn))
1248                                         d_rehash(dn);
1249                                 update_dentry_lease(dn, rinfo->dlease,
1250                                                     session,
1251                                                     req->r_request_started);
1252                         }
1253                         goto done;
1254                 }
1255
1256                 /* attach proper inode */
1257                 if (!dn->d_inode) {
1258                         ceph_dir_clear_ordered(dir);
1259                         ihold(in);
1260                         dn = splice_dentry(dn, in, &have_lease);
1261                         if (IS_ERR(dn)) {
1262                                 err = PTR_ERR(dn);
1263                                 goto done;
1264                         }
1265                         req->r_dentry = dn;  /* may have spliced */
1266                 } else if (dn->d_inode && dn->d_inode != in) {
1267                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1268                              dn, dn->d_inode, ceph_vinop(dn->d_inode),
1269                              ceph_vinop(in));
1270                         have_lease = false;
1271                 }
1272
1273                 if (have_lease)
1274                         update_dentry_lease(dn, rinfo->dlease, session,
1275                                             req->r_request_started);
1276                 dout(" final dn %p\n", dn);
1277         } else if (!req->r_aborted &&
1278                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1279                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1280                 struct dentry *dn = req->r_dentry;
1281                 struct inode *dir = req->r_locked_dir;
1282
1283                 /* fill out a snapdir LOOKUPSNAP dentry */
1284                 BUG_ON(!dn);
1285                 BUG_ON(!dir);
1286                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1287                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1288                 ceph_dir_clear_ordered(dir);
1289                 ihold(in);
1290                 dn = splice_dentry(dn, in, NULL);
1291                 if (IS_ERR(dn)) {
1292                         err = PTR_ERR(dn);
1293                         goto done;
1294                 }
1295                 req->r_dentry = dn;  /* may have spliced */
1296         }
1297 done:
1298         dout("fill_trace done err=%d\n", err);
1299         return err;
1300 }
1301
1302 /*
1303  * Prepopulate our cache with readdir results, leases, etc.
1304  */
1305 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1306                                            struct ceph_mds_session *session)
1307 {
1308         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1309         int i, err = 0;
1310
1311         for (i = 0; i < rinfo->dir_nr; i++) {
1312                 struct ceph_vino vino;
1313                 struct inode *in;
1314                 int rc;
1315
1316                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1317                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1318
1319                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1320                 if (IS_ERR(in)) {
1321                         err = PTR_ERR(in);
1322                         dout("new_inode badness got %d\n", err);
1323                         continue;
1324                 }
1325                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1326                                 req->r_request_started, -1,
1327                                 &req->r_caps_reservation);
1328                 if (rc < 0) {
1329                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1330                         err = rc;
1331                         continue;
1332                 }
1333         }
1334
1335         return err;
1336 }
1337
1338 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1339                              struct ceph_mds_session *session)
1340 {
1341         struct dentry *parent = req->r_dentry;
1342         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1343         struct qstr dname;
1344         struct dentry *dn;
1345         struct inode *in;
1346         int err = 0, ret, i;
1347         struct inode *snapdir = NULL;
1348         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1349         struct ceph_dentry_info *di;
1350         u64 r_readdir_offset = req->r_readdir_offset;
1351         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1352
1353         if (rinfo->dir_dir &&
1354             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1355                 dout("readdir_prepopulate got new frag %x -> %x\n",
1356                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1357                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1358                 if (ceph_frag_is_leftmost(frag))
1359                         r_readdir_offset = 2;
1360                 else
1361                         r_readdir_offset = 0;
1362         }
1363
1364         if (req->r_aborted)
1365                 return readdir_prepopulate_inodes_only(req, session);
1366
1367         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1368                 snapdir = ceph_get_snapdir(parent->d_inode);
1369                 parent = d_find_alias(snapdir);
1370                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1371                      rinfo->dir_nr, parent);
1372         } else {
1373                 dout("readdir_prepopulate %d items under dn %p\n",
1374                      rinfo->dir_nr, parent);
1375                 if (rinfo->dir_dir)
1376                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1377         }
1378
1379         /* FIXME: release caps/leases if error occurs */
1380         for (i = 0; i < rinfo->dir_nr; i++) {
1381                 struct ceph_vino vino;
1382
1383                 dname.name = rinfo->dir_dname[i];
1384                 dname.len = rinfo->dir_dname_len[i];
1385                 dname.hash = full_name_hash(dname.name, dname.len);
1386
1387                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1388                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1389
1390 retry_lookup:
1391                 dn = d_lookup(parent, &dname);
1392                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1393                      parent, dname.len, dname.name, dn);
1394
1395                 if (!dn) {
1396                         dn = d_alloc(parent, &dname);
1397                         dout("d_alloc %p '%.*s' = %p\n", parent,
1398                              dname.len, dname.name, dn);
1399                         if (dn == NULL) {
1400                                 dout("d_alloc badness\n");
1401                                 err = -ENOMEM;
1402                                 goto out;
1403                         }
1404                         ret = ceph_init_dentry(dn);
1405                         if (ret < 0) {
1406                                 dput(dn);
1407                                 err = ret;
1408                                 goto out;
1409                         }
1410                 } else if (dn->d_inode &&
1411                            (ceph_ino(dn->d_inode) != vino.ino ||
1412                             ceph_snap(dn->d_inode) != vino.snap)) {
1413                         dout(" dn %p points to wrong inode %p\n",
1414                              dn, dn->d_inode);
1415                         d_delete(dn);
1416                         dput(dn);
1417                         goto retry_lookup;
1418                 } else {
1419                         /* reorder parent's d_subdirs */
1420                         spin_lock(&parent->d_lock);
1421                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1422                         list_move(&dn->d_child, &parent->d_subdirs);
1423                         spin_unlock(&dn->d_lock);
1424                         spin_unlock(&parent->d_lock);
1425                 }
1426
1427                 /* inode */
1428                 if (dn->d_inode) {
1429                         in = dn->d_inode;
1430                 } else {
1431                         in = ceph_get_inode(parent->d_sb, vino);
1432                         if (IS_ERR(in)) {
1433                                 dout("new_inode badness\n");
1434                                 d_drop(dn);
1435                                 dput(dn);
1436                                 err = PTR_ERR(in);
1437                                 goto out;
1438                         }
1439                 }
1440
1441                 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1442                                req->r_request_started, -1,
1443                                &req->r_caps_reservation) < 0) {
1444                         pr_err("fill_inode badness on %p\n", in);
1445                         if (!dn->d_inode)
1446                                 iput(in);
1447                         d_drop(dn);
1448                         goto next_item;
1449                 }
1450
1451                 if (!dn->d_inode) {
1452                         struct dentry *realdn = splice_dentry(dn, in, NULL);
1453                         if (IS_ERR(realdn)) {
1454                                 err = PTR_ERR(realdn);
1455                                 d_drop(dn);
1456                                 dn = NULL;
1457                                 goto next_item;
1458                         }
1459                         dn = realdn;
1460                 }
1461
1462                 di = dn->d_fsdata;
1463                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1464
1465                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1466                                     req->r_session,
1467                                     req->r_request_started);
1468 next_item:
1469                 if (dn)
1470                         dput(dn);
1471         }
1472         if (err == 0)
1473                 req->r_did_prepopulate = true;
1474
1475 out:
1476         if (snapdir) {
1477                 iput(snapdir);
1478                 dput(parent);
1479         }
1480         dout("readdir_prepopulate done\n");
1481         return err;
1482 }
1483
1484 int ceph_inode_set_size(struct inode *inode, loff_t size)
1485 {
1486         struct ceph_inode_info *ci = ceph_inode(inode);
1487         int ret = 0;
1488
1489         spin_lock(&ci->i_ceph_lock);
1490         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1491         inode->i_size = size;
1492         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1493
1494         /* tell the MDS if we are approaching max_size */
1495         if ((size << 1) >= ci->i_max_size &&
1496             (ci->i_reported_size << 1) < ci->i_max_size)
1497                 ret = 1;
1498
1499         spin_unlock(&ci->i_ceph_lock);
1500         return ret;
1501 }
1502
1503 /*
1504  * Write back inode data in a worker thread.  (This can't be done
1505  * in the message handler context.)
1506  */
1507 void ceph_queue_writeback(struct inode *inode)
1508 {
1509         ihold(inode);
1510         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1511                        &ceph_inode(inode)->i_wb_work)) {
1512                 dout("ceph_queue_writeback %p\n", inode);
1513         } else {
1514                 dout("ceph_queue_writeback %p failed\n", inode);
1515                 iput(inode);
1516         }
1517 }
1518
1519 static void ceph_writeback_work(struct work_struct *work)
1520 {
1521         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1522                                                   i_wb_work);
1523         struct inode *inode = &ci->vfs_inode;
1524
1525         dout("writeback %p\n", inode);
1526         filemap_fdatawrite(&inode->i_data);
1527         iput(inode);
1528 }
1529
1530 /*
1531  * queue an async invalidation
1532  */
1533 void ceph_queue_invalidate(struct inode *inode)
1534 {
1535         ihold(inode);
1536         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1537                        &ceph_inode(inode)->i_pg_inv_work)) {
1538                 dout("ceph_queue_invalidate %p\n", inode);
1539         } else {
1540                 dout("ceph_queue_invalidate %p failed\n", inode);
1541                 iput(inode);
1542         }
1543 }
1544
1545 /*
1546  * Invalidate inode pages in a worker thread.  (This can't be done
1547  * in the message handler context.)
1548  */
1549 static void ceph_invalidate_work(struct work_struct *work)
1550 {
1551         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1552                                                   i_pg_inv_work);
1553         struct inode *inode = &ci->vfs_inode;
1554         u32 orig_gen;
1555         int check = 0;
1556
1557         mutex_lock(&ci->i_truncate_mutex);
1558         spin_lock(&ci->i_ceph_lock);
1559         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1560              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1561         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1562                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1563                         check = 1;
1564                 spin_unlock(&ci->i_ceph_lock);
1565                 mutex_unlock(&ci->i_truncate_mutex);
1566                 goto out;
1567         }
1568         orig_gen = ci->i_rdcache_gen;
1569         spin_unlock(&ci->i_ceph_lock);
1570
1571         truncate_pagecache(inode, 0);
1572
1573         spin_lock(&ci->i_ceph_lock);
1574         if (orig_gen == ci->i_rdcache_gen &&
1575             orig_gen == ci->i_rdcache_revoking) {
1576                 dout("invalidate_pages %p gen %d successful\n", inode,
1577                      ci->i_rdcache_gen);
1578                 ci->i_rdcache_revoking--;
1579                 check = 1;
1580         } else {
1581                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1582                      inode, orig_gen, ci->i_rdcache_gen,
1583                      ci->i_rdcache_revoking);
1584                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1585                         check = 1;
1586         }
1587         spin_unlock(&ci->i_ceph_lock);
1588         mutex_unlock(&ci->i_truncate_mutex);
1589 out:
1590         if (check)
1591                 ceph_check_caps(ci, 0, NULL);
1592         iput(inode);
1593 }
1594
1595
1596 /*
1597  * called by trunc_wq;
1598  *
1599  * We also truncate in a separate thread as well.
1600  */
1601 static void ceph_vmtruncate_work(struct work_struct *work)
1602 {
1603         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1604                                                   i_vmtruncate_work);
1605         struct inode *inode = &ci->vfs_inode;
1606
1607         dout("vmtruncate_work %p\n", inode);
1608         __ceph_do_pending_vmtruncate(inode);
1609         iput(inode);
1610 }
1611
1612 /*
1613  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1614  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1615  */
1616 void ceph_queue_vmtruncate(struct inode *inode)
1617 {
1618         struct ceph_inode_info *ci = ceph_inode(inode);
1619
1620         ihold(inode);
1621
1622         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1623                        &ci->i_vmtruncate_work)) {
1624                 dout("ceph_queue_vmtruncate %p\n", inode);
1625         } else {
1626                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1627                      inode, ci->i_truncate_pending);
1628                 iput(inode);
1629         }
1630 }
1631
1632 /*
1633  * Make sure any pending truncation is applied before doing anything
1634  * that may depend on it.
1635  */
1636 void __ceph_do_pending_vmtruncate(struct inode *inode)
1637 {
1638         struct ceph_inode_info *ci = ceph_inode(inode);
1639         u64 to;
1640         int wrbuffer_refs, finish = 0;
1641
1642         mutex_lock(&ci->i_truncate_mutex);
1643 retry:
1644         spin_lock(&ci->i_ceph_lock);
1645         if (ci->i_truncate_pending == 0) {
1646                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1647                 spin_unlock(&ci->i_ceph_lock);
1648                 mutex_unlock(&ci->i_truncate_mutex);
1649                 return;
1650         }
1651
1652         /*
1653          * make sure any dirty snapped pages are flushed before we
1654          * possibly truncate them.. so write AND block!
1655          */
1656         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1657                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1658                      inode);
1659                 spin_unlock(&ci->i_ceph_lock);
1660                 filemap_write_and_wait_range(&inode->i_data, 0,
1661                                              inode->i_sb->s_maxbytes);
1662                 goto retry;
1663         }
1664
1665         /* there should be no reader or writer */
1666         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1667
1668         to = ci->i_truncate_size;
1669         wrbuffer_refs = ci->i_wrbuffer_ref;
1670         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1671              ci->i_truncate_pending, to);
1672         spin_unlock(&ci->i_ceph_lock);
1673
1674         truncate_pagecache(inode, to);
1675
1676         spin_lock(&ci->i_ceph_lock);
1677         if (to == ci->i_truncate_size) {
1678                 ci->i_truncate_pending = 0;
1679                 finish = 1;
1680         }
1681         spin_unlock(&ci->i_ceph_lock);
1682         if (!finish)
1683                 goto retry;
1684
1685         mutex_unlock(&ci->i_truncate_mutex);
1686
1687         if (wrbuffer_refs == 0)
1688                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1689
1690         wake_up_all(&ci->i_cap_wq);
1691 }
1692
1693 /*
1694  * symlinks
1695  */
1696 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1697 {
1698         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1699         nd_set_link(nd, ci->i_symlink);
1700         return NULL;
1701 }
1702
1703 static const struct inode_operations ceph_symlink_iops = {
1704         .readlink = generic_readlink,
1705         .follow_link = ceph_sym_follow_link,
1706         .setattr = ceph_setattr,
1707         .getattr = ceph_getattr,
1708         .setxattr = ceph_setxattr,
1709         .getxattr = ceph_getxattr,
1710         .listxattr = ceph_listxattr,
1711         .removexattr = ceph_removexattr,
1712 };
1713
1714 /*
1715  * setattr
1716  */
1717 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1718 {
1719         struct inode *inode = dentry->d_inode;
1720         struct ceph_inode_info *ci = ceph_inode(inode);
1721         const unsigned int ia_valid = attr->ia_valid;
1722         struct ceph_mds_request *req;
1723         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1724         int issued;
1725         int release = 0, dirtied = 0;
1726         int mask = 0;
1727         int err = 0;
1728         int inode_dirty_flags = 0;
1729
1730         if (ceph_snap(inode) != CEPH_NOSNAP)
1731                 return -EROFS;
1732
1733         err = inode_change_ok(inode, attr);
1734         if (err != 0)
1735                 return err;
1736
1737         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1738                                        USE_AUTH_MDS);
1739         if (IS_ERR(req))
1740                 return PTR_ERR(req);
1741
1742         spin_lock(&ci->i_ceph_lock);
1743         issued = __ceph_caps_issued(ci, NULL);
1744         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1745
1746         if (ia_valid & ATTR_UID) {
1747                 dout("setattr %p uid %d -> %d\n", inode,
1748                      from_kuid(&init_user_ns, inode->i_uid),
1749                      from_kuid(&init_user_ns, attr->ia_uid));
1750                 if (issued & CEPH_CAP_AUTH_EXCL) {
1751                         inode->i_uid = attr->ia_uid;
1752                         dirtied |= CEPH_CAP_AUTH_EXCL;
1753                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1754                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1755                         req->r_args.setattr.uid = cpu_to_le32(
1756                                 from_kuid(&init_user_ns, attr->ia_uid));
1757                         mask |= CEPH_SETATTR_UID;
1758                         release |= CEPH_CAP_AUTH_SHARED;
1759                 }
1760         }
1761         if (ia_valid & ATTR_GID) {
1762                 dout("setattr %p gid %d -> %d\n", inode,
1763                      from_kgid(&init_user_ns, inode->i_gid),
1764                      from_kgid(&init_user_ns, attr->ia_gid));
1765                 if (issued & CEPH_CAP_AUTH_EXCL) {
1766                         inode->i_gid = attr->ia_gid;
1767                         dirtied |= CEPH_CAP_AUTH_EXCL;
1768                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1769                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1770                         req->r_args.setattr.gid = cpu_to_le32(
1771                                 from_kgid(&init_user_ns, attr->ia_gid));
1772                         mask |= CEPH_SETATTR_GID;
1773                         release |= CEPH_CAP_AUTH_SHARED;
1774                 }
1775         }
1776         if (ia_valid & ATTR_MODE) {
1777                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1778                      attr->ia_mode);
1779                 if (issued & CEPH_CAP_AUTH_EXCL) {
1780                         inode->i_mode = attr->ia_mode;
1781                         dirtied |= CEPH_CAP_AUTH_EXCL;
1782                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1783                            attr->ia_mode != inode->i_mode) {
1784                         inode->i_mode = attr->ia_mode;
1785                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1786                         mask |= CEPH_SETATTR_MODE;
1787                         release |= CEPH_CAP_AUTH_SHARED;
1788                 }
1789         }
1790
1791         if (ia_valid & ATTR_ATIME) {
1792                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1793                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1794                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1795                 if (issued & CEPH_CAP_FILE_EXCL) {
1796                         ci->i_time_warp_seq++;
1797                         inode->i_atime = attr->ia_atime;
1798                         dirtied |= CEPH_CAP_FILE_EXCL;
1799                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1800                            timespec_compare(&inode->i_atime,
1801                                             &attr->ia_atime) < 0) {
1802                         inode->i_atime = attr->ia_atime;
1803                         dirtied |= CEPH_CAP_FILE_WR;
1804                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1805                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1806                         ceph_encode_timespec(&req->r_args.setattr.atime,
1807                                              &attr->ia_atime);
1808                         mask |= CEPH_SETATTR_ATIME;
1809                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1810                                 CEPH_CAP_FILE_WR;
1811                 }
1812         }
1813         if (ia_valid & ATTR_MTIME) {
1814                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1815                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1816                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1817                 if (issued & CEPH_CAP_FILE_EXCL) {
1818                         ci->i_time_warp_seq++;
1819                         inode->i_mtime = attr->ia_mtime;
1820                         dirtied |= CEPH_CAP_FILE_EXCL;
1821                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1822                            timespec_compare(&inode->i_mtime,
1823                                             &attr->ia_mtime) < 0) {
1824                         inode->i_mtime = attr->ia_mtime;
1825                         dirtied |= CEPH_CAP_FILE_WR;
1826                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1827                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1828                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1829                                              &attr->ia_mtime);
1830                         mask |= CEPH_SETATTR_MTIME;
1831                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1832                                 CEPH_CAP_FILE_WR;
1833                 }
1834         }
1835         if (ia_valid & ATTR_SIZE) {
1836                 dout("setattr %p size %lld -> %lld\n", inode,
1837                      inode->i_size, attr->ia_size);
1838                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1839                     attr->ia_size > inode->i_size) {
1840                         inode->i_size = attr->ia_size;
1841                         inode->i_blocks =
1842                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1843                         inode->i_ctime = attr->ia_ctime;
1844                         ci->i_reported_size = attr->ia_size;
1845                         dirtied |= CEPH_CAP_FILE_EXCL;
1846                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1847                            attr->ia_size != inode->i_size) {
1848                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1849                         req->r_args.setattr.old_size =
1850                                 cpu_to_le64(inode->i_size);
1851                         mask |= CEPH_SETATTR_SIZE;
1852                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1853                                 CEPH_CAP_FILE_WR;
1854                 }
1855         }
1856
1857         /* these do nothing */
1858         if (ia_valid & ATTR_CTIME) {
1859                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1860                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1861                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1862                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1863                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1864                      only ? "ctime only" : "ignored");
1865                 inode->i_ctime = attr->ia_ctime;
1866                 if (only) {
1867                         /*
1868                          * if kernel wants to dirty ctime but nothing else,
1869                          * we need to choose a cap to dirty under, or do
1870                          * a almost-no-op setattr
1871                          */
1872                         if (issued & CEPH_CAP_AUTH_EXCL)
1873                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1874                         else if (issued & CEPH_CAP_FILE_EXCL)
1875                                 dirtied |= CEPH_CAP_FILE_EXCL;
1876                         else if (issued & CEPH_CAP_XATTR_EXCL)
1877                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1878                         else
1879                                 mask |= CEPH_SETATTR_CTIME;
1880                 }
1881         }
1882         if (ia_valid & ATTR_FILE)
1883                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1884
1885         if (dirtied) {
1886                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1887                 inode->i_ctime = CURRENT_TIME;
1888         }
1889
1890         release &= issued;
1891         spin_unlock(&ci->i_ceph_lock);
1892
1893         if (inode_dirty_flags)
1894                 __mark_inode_dirty(inode, inode_dirty_flags);
1895
1896         if (ia_valid & ATTR_MODE) {
1897                 err = posix_acl_chmod(inode, attr->ia_mode);
1898                 if (err)
1899                         goto out_put;
1900         }
1901
1902         if (mask) {
1903                 req->r_inode = inode;
1904                 ihold(inode);
1905                 req->r_inode_drop = release;
1906                 req->r_args.setattr.mask = cpu_to_le32(mask);
1907                 req->r_num_caps = 1;
1908                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1909         }
1910         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1911              ceph_cap_string(dirtied), mask);
1912
1913         ceph_mdsc_put_request(req);
1914         if (mask & CEPH_SETATTR_SIZE)
1915                 __ceph_do_pending_vmtruncate(inode);
1916         return err;
1917 out_put:
1918         ceph_mdsc_put_request(req);
1919         return err;
1920 }
1921
1922 /*
1923  * Verify that we have a lease on the given mask.  If not,
1924  * do a getattr against an mds.
1925  */
1926 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1927                       int mask, bool force)
1928 {
1929         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1930         struct ceph_mds_client *mdsc = fsc->mdsc;
1931         struct ceph_mds_request *req;
1932         int err;
1933
1934         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1935                 dout("do_getattr inode %p SNAPDIR\n", inode);
1936                 return 0;
1937         }
1938
1939         dout("do_getattr inode %p mask %s mode 0%o\n",
1940              inode, ceph_cap_string(mask), inode->i_mode);
1941         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1942                 return 0;
1943
1944         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1945         if (IS_ERR(req))
1946                 return PTR_ERR(req);
1947         req->r_inode = inode;
1948         ihold(inode);
1949         req->r_num_caps = 1;
1950         req->r_args.getattr.mask = cpu_to_le32(mask);
1951         req->r_locked_page = locked_page;
1952         err = ceph_mdsc_do_request(mdsc, NULL, req);
1953         if (locked_page && err == 0) {
1954                 u64 inline_version = req->r_reply_info.targeti.inline_version;
1955                 if (inline_version == 0) {
1956                         /* the reply is supposed to contain inline data */
1957                         err = -EINVAL;
1958                 } else if (inline_version == CEPH_INLINE_NONE) {
1959                         err = -ENODATA;
1960                 } else {
1961                         err = req->r_reply_info.targeti.inline_len;
1962                 }
1963         }
1964         ceph_mdsc_put_request(req);
1965         dout("do_getattr result=%d\n", err);
1966         return err;
1967 }
1968
1969
1970 /*
1971  * Check inode permissions.  We verify we have a valid value for
1972  * the AUTH cap, then call the generic handler.
1973  */
1974 int ceph_permission(struct inode *inode, int mask)
1975 {
1976         int err;
1977
1978         if (mask & MAY_NOT_BLOCK)
1979                 return -ECHILD;
1980
1981         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1982
1983         if (!err)
1984                 err = generic_permission(inode, mask);
1985         return err;
1986 }
1987
1988 /*
1989  * Get all attributes.  Hopefully somedata we'll have a statlite()
1990  * and can limit the fields we require to be accurate.
1991  */
1992 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1993                  struct kstat *stat)
1994 {
1995         struct inode *inode = dentry->d_inode;
1996         struct ceph_inode_info *ci = ceph_inode(inode);
1997         int err;
1998
1999         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2000         if (!err) {
2001                 generic_fillattr(inode, stat);
2002                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2003                 if (ceph_snap(inode) != CEPH_NOSNAP)
2004                         stat->dev = ceph_snap(inode);
2005                 else
2006                         stat->dev = 0;
2007                 if (S_ISDIR(inode->i_mode)) {
2008                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2009                                                 RBYTES))
2010                                 stat->size = ci->i_rbytes;
2011                         else
2012                                 stat->size = ci->i_files + ci->i_subdirs;
2013                         stat->blocks = 0;
2014                         stat->blksize = 65536;
2015                 }
2016         }
2017         return err;
2018 }